macOS · Apple Silicon · on-device

Cloud-quality transcription, fully local.

Press a key and speak. Clean, formatted text appears wherever your cursor is, transcribed and polished entirely on your Mac. No cloud, no account, from $2 a month or $20 to own it.

100% on-device From $2/mo $20 to own it

Free trial: 3,000 words or 7 days. No account, no card.

Prefer the terminal?
brew install --cask evoglyph/tap/evoglyph
curl -fsSL https://evoglyph.com/install.sh | sh
~0.6s
speech to text
100%
on-device
4x
faster than typing
$20
own it forever

Full pipeline, p50, short dictation on an Apple M4 Max. Scales with length. Full method on the benchmark.

On-device cleanup

Raw speech in. Clean text out.

Injecting a raw transcript into Slack means signing your name to your unedited inner monologue. evoglyph fixes it first, on-device.

Recording

What you said · raw transcript

um so yeah i think we should uh we should probably ship the the parakeet fix today because like the latency thing is is basically gone now you know

Cleaned by evoglyph

What lands · injected into Slack

I think we should ship the Parakeet fix today. The latency regression is basically gone now.

LFM2-2.6B (4-bit) running in-process. No server, no network call.

~0.6s speech to finished text,
one-sentence dictation (~10 words)

On an Apple M4 Max, a short one-sentence dictation goes from speech to finished, cleaned-up text in about 0.6 seconds. That is transcription and the AI cleanup pass together, 100% on-device. About 95% of short dictations finish in under 1.5 seconds.

~73% under 1s ~95% under 1.5s transcription stage alone ~0.3s

Full pipeline (transcription + AI cleanup), measured across ~2,500 real dictations on an Apple M4 Max. Latency scales with how long you speak, so longer dictations take longer. Slower Apple Silicon is unmeasured and will be higher. Full method on the benchmark.

How it works

Three steps, then four reasons.

Press a key, speak, and clean text lands in whatever app you are in. Everything runs on your Mac, built for the way developers actually write.

01

Press

Hit your hotkey from any app. Global capture works system-wide, with no window to open.

02

Speak

Talk naturally at around 180 words a minute. Parakeet runs on the Apple Neural Engine, on-device.

03

Done

Clean, polished text appears under your cursor in about 0.6 seconds, across 16+ apps.

Fully local

Audio and transcripts physically never leave the Mac. No cloud transcription service exists on the backend.

Fast, on-device

Speech becomes finished, cleaned-up text in about 0.6 seconds, transcription and an on-device AI cleanup pass, with no server round trip.

Works everywhere

Verified across 16+ apps. It adapts its injection method per app, so native fields, browsers, and terminals all just work.

Built for developers

One native Swift binary near 150 MB, not an 800 MB Electron app. Vocabulary boosting gets your technical terms right.

Verified in 16+ apps

Slack VS Code Cursor Chrome Safari Obsidian Notion Linear Terminal Claude ChatGPT

The actual moat

Fully on your Mac, and a fraction of the cost.

There is no cloud bill on our side, so the price stays low: $2 a month, $10 a year, or $20 once to own it. Cloud tools charge up to $144 a year because every word costs them GPU time.

$2/mo · $10/yr · $20 once

See pricing →
Example copy · replace with real quotes before publishing

Built for people who write all day.

"I dictate every commit message and PR description now. It lands in under a second and the cleanup means I can ramble and still get clean prose. Nothing leaves my laptop."

[Name]

Staff engineer, [Company]

"I tried the cloud tools and the lag broke my flow every time. evoglyph is instant and I trust it with client work because the audio never goes anywhere."

[Name]

Indie developer, [Product]

"Vocabulary boosting nails our product names and Kubernetes terms on the first try. It replaced a $144/yr subscription for me and runs faster."

[Name]

Platform lead, [Company]

How it compares

Fully on your device, and yours to keep.

evoglyph's edge is not a single speed number, so the table does not lead with one. It leads with what a cloud subscription structurally cannot offer: everything runs on your device, and you can own it for a one-time price.

Feature
evoglyph
Wispr Flow
Superwhisper
OpenWhispr
On-device transcription
✓Yes
✕No
✓Yes
✓Yes
On-device AI cleanup (default)
✓Yes
Cloud
Optional
Optional
One-time pricing option
$20
✕No
$249
Free
Cross-platform (beyond macOS)
✕No
✓Yes
✓Yes
✓Yes
Native binary, not Electron
~150 MB
~500 MB
✓Yes
✕No

Where they win: Wispr Flow covers Windows, iOS, and Android with team and compliance programs (SOC 2, HIPAA) evoglyph does not. Superwhisper and OpenWhispr also run locally and reach beyond macOS. If you need cross-platform or formal compliance today, those are honest picks. evoglyph's bet is the opposite: one focused, native macOS app, with on-device cleanup on by default and a one-time price. Competitor cells as of June 2026, from each tool's official site; re-verified before each publish.

Pricing

Priced to read as a steal, not as cheap.

Cloud dictation elsewhere: Wispr Flow $144/yr · Superwhisper $249. evoglyph: $10/yr or $20 once.

Monthly

Cancel anytime.

$2 /month
  • All local features
  • On-device cleanup
  • Free updates while subscribed
Start the trial Subscribe now
Most popular

Annual

Save 58% vs monthly.

$10 /year
Save 58%
  • Everything in Monthly
  • Priority support
  • Best value for daily use
Start the trial Subscribe now

Lifetime

Pay once, yours forever.

$20 once
  • All features, forever
  • All future updates
  • No recurring charge
Start the trial Buy lifetime

Why is it this cheap?

evoglyph has no cloud bill, so there is nothing to pass on to you. Cloud tools charge $144 a year because every word you speak costs them GPU time. The trial is 3,000 words or 7 days, no account and no card.

Early access

Have an invite code? Start the free trial; your cohort discount is applied at checkout when you upgrade.

Start the trial

Privacy

Your voice stays on your Mac.

Six claims, each with a one-line way to confirm it yourself. This audience checks, so every privacy claim ships with a way to verify it.

Local transcription only

Speech is transcribed on your Mac. No transcription service exists on the backend.

verify: watch the network tab while dictating

On-device processing

Both the speech model and the cleanup LLM run on the Apple Neural Engine, on your machine.

verify: pull the network cable, it still works

First-party crash reporting

Crash reports are opt-in and carry no audio and no transcripts, ever. No third-party telemetry vendor.

verify: read the disclosed payload

Disk safety

History lives in a local database you can erase at any time from Settings. Audio buffers are processed in memory.

verify: inspect the local SQLite file

Every network call listed

Model download, license check, update check, opt-in diagnostics. Never audio or transcripts. See the full disclosure.

verify: run lsof -i on the process

Menu-bar daemon

A small native Swift binary, not a window you open or an Electron shell. No dock icon.

verify: check Activity Monitor for RAM

Models

Speech recognition
NVIDIA Parakeet TDT, on the Apple Neural Engine. 2.41% word error rate.
Text cleanup
Liquid AI LFM2-2.6B (4-bit), running in-process. Cleans up and polishes what you said, on-device.
First launch
A one-time model download (about 4 GB), then fully offline.

Pipeline

  1. 1

    Capture audio on your hotkey.

  2. 2

    Transcribe on the Apple Neural Engine.

  3. 3

    Clean in-process with the LLM.

  4. 4

    Inject into the focused app.

Transcribe, clean, and inject happen in well under a second for normal phrases. Every stage runs on-device.

Questions

Frequently asked.

An Apple Silicon Mac (M1 or later) on macOS 14 or newer. The models download once on first launch (about 4 GB), then it runs fully offline. AI cleanup is built in, with no separate runtime to install.

View all FAQs

Start dictating in minutes.

Press a key, speak naturally, and clean text appears wherever your cursor is. Fully on your Mac.

Download for macOS See pricing

Free trial: 3,000 words or 7 days. No account, no card.